A study on the epidemiology of incident seizures in patients with neuropsychiatric disorders

(1)

A study on the epidemiology of incident seizures in patients with neuropsychiatric

disorders

Inauguraldissertation zur

Erlangung der Würde eines Doktors der Philosophie vorgelegt der

Philosophisch-Naturwissenschaftlichen Fakultät der Universität Basel

von

Marlene Susanne Blöchliger

aus Ernetschwil (SG)

Basel, 2015

Originaldokument gespeichert auf dem Dokumentenserver der Universität

Basel edoc.unibas.ch

(2)

Genehmigt von der Philosophisch-Naturwissenschaftlichen Fakultät auf Antrag von

Prof. Dr. Christoph Meier

Prof. Dr. Dr. Stephan Krähenbühl

Basel, den 15.September 2015

Prof. Dr. Jörg Schibler

Dekan

(3)

The various risk factors for seizures, a painting by Franziska Rauch (2015)

(4)

(5)

Acknowledgments

This PhD thesis was the team play of many people who I would like to thank here.

I am especially grateful to my thesis supervisor Prof. Dr. Christoph Meier, who has given me the chance to conduct my PhD thesis at his group. Thank you very much Christoph for your considerable confidence in me and my work, your continuous optimism, and your generosity.

I enjoyed all the freedom I needed to pursue this thesis in an independent manner and learn about epidemiology. I am very happy that I can continue working at the BPU a bit longer, and look forward to what is coming next.

I would also like to thank my project supervisor PD Dr. Michael Bodmer, who supported my projects with a research grant, had very creative ideas on what to investigate, and helped me with the planning and conducting of the research, and the editing of the manuscripts. Thank you Michael, for having given me all the freedom I needed to pursue my own goals, try out new things, and for having made me realize that I can manage a lot of things pretty much on my own.

Further thanks go to PD Dr. Stephan Rüegg, for your genuine interest in my thesis projects, for proof-reading protocols and manuscripts, for sharing your clinical knowledge, and for co- authoring our manuscripts in such a supportive and motivating manner.

Also, I would like to thank Prof. Dr. Dr. Stephan Krähenbühl for being the second examiner, and Prof. Dr. Henriette Meyer zu Schwabedissen for acting as the chair at the defense.

I would like to thank Prof. Susan Jick from the Boston Collaborative Drug Surveillance Program (BCDSP), for proof-reading and co-authoring our manuscripts. Thank you very much Sue, for your critical appraisal of my work, and also that you would have given me the opportunity to pursue part of my thesis at the BCDSP.

Many thanks go to my colleagues of the Basel Pharmacoepidemiology Unit (BPU), namely Pascal Egger, Dr. Cornelia Schneider, Nadja Stohler, Dr. Daphne Reinau, Noel Frey, Fabienne Bietry, Dr. Claire Wilson, Dr. Saskia Bruderer, Dr. Julia Spöndlin, Dr. Claudia Becker, Dr.

Patrick Imfeld, Alexandra Müller, and Delia Bornand, who have always supported me when I had questions, and who are or were dear desk neighbors and companions for coffee breaks, lunches, and many social events.

(6)

Thank you Pascal, for the great job with the programming of my studies, for your substantial stress resistance, and for your considerable patience when it comes to unclearly worded requests or when everyone needs data immediately.

Daniela and Karen, thank you for your extremely positive attitude and kindness, meeting you has always been a highlight. Simone, thanks for having stuck around over the years, and for your refreshing humor and pragmatism. Carole, thank you for mutual motivation and encouragement towards the end of the PhD. Julia, thank you for having supported me with my projects, the many nice apéros, dinners, and city trips we shared, and especially holy sisterhood. Saskia, thank for having taught SAS to me when I urgently needed help, for having shared everything from sports sessions to hotel rooms to LaTex templates, and, above all, for having walked along at the high and at the low times. Oli, Regi, and Barbara, thanks so much for your unconditional friendship throughout the last decades.

Special thanks go to my parents Ursula and Peter. All along you have supported me, my brother, and our families, in every way you could. You are really great, and I hope you realize how much it means to have parents like you.

Many thanks also go to Alex’s parents, Franziska and Roger. Thank you both so much for your generosity and support, and all the wonderful places we have visited as the ever growing Rudi’s family. Franziska, thank you for travelling to Olten all these years to look after Enea, and for the amazing picture you drew for my PhD thesis.

Last but not least, thank you Enea, for accepting that in the past few years I did not always have enough time for you, and for reminding me that I did not have to fear speaking in front of big audiences. Above all, thank you so much Alex, for proof-reading my thesis, fiddling around with my figures and presentations until they looked nice, believing in me, being patient, and above all, being as solid as a rock.

(7)

Table of Contents

I. Summary ... i

II.Abbreviations ... iii

1. Introduction ... 1

1.1 Pharmacoepidemiology ... 1

1.1.1 Defining pharmacoepidemiology ... 1

1.1.2 Historical development of pharmacoepidemiologic studies ... 1

1.1.3 Clinical trials and the role of pharmacoepidemiology ... 2

1.1.3.1 Clinical trials: phase I to III ... 2

1.1.3.2 Clinical trials: phase IV ... 3

1.1.4 Association types in pharmacoepidemiology ... 3

1.1.4.1 No association ... 4

1.1.4.2 Chance and bias – two types of artifactual associations ... 4

1.1.4.3 Indirect associations ... 5

1.1.4.4 Causal associations ... 6

1.1.5 Study designs in pharmacoepidemiology ... 7

1.1.5.1 Observational studies ... 7

1.1.5.1.1 Descriptive studies ... 7

1.1.5.1.2 Analytical studies ... 8

1.1.5.2 Experimental (interventional) studies ... 13

1.1.6 Data sources in pharmacoepidemiology ... 14

1.2 Seizures and Epilepsy ... 16

1.2.1 Defining seizures and epilepsy ... 16

1.2.2 History of seizures and epilepsy ... 16

1.2.3 Epidemiology of seizures and epilepsy ... 17

1.2.4 Risk factors for seizures and epilepsy ... 18

1.2.5 Mechanism of seizures ... 19

1.2.6 Seizure types ... 19

1.2.7 Diagnosis of seizures ... 21

1.2.8 Therapy of seizures ... 22

2. Aims of the thesis ... 25

3. Seizure projects ... 29

3.1 Project 1: Risk factors for seizures in adult patients with depression ... 29

3.1.1 Abstract ... 30

3.1.2 Introduction... 31

3.1.3 Methods ... 31

3.1.4 Results ... 34

(8)

3.1.5 Discussion ... 44

3.2 Project 2: Antidepressant drug use and the risk of seizures ... 47

3.2.1 Abstract ... 48

3.2.3 Methods ... 49

3.2.4 Results ... 53

3.3 Project 3: Antipsychotic drug use and the risk of seizures ... 75

3.3.1 Abstract ... 76

3.3.3 Methods ... 78

3.3.4 Results ... 83

4. Discussion ... 105

4.1 Strengths and limitations of the data source ... 106

4.1.1 Sample size limitations ... 106

4.1.2 Validity of drug exposures and medical diagnoses ... 107

4.1.2.1 Validity of drug exposures ... 107

4.1.2.2 General validity of medical diagnoses ... 109

4.1.2.3 Validity of seizure diagnosis ... 110

4.2 Strengths and limitations of the observational approach ... 111

4.2.1 Chance ... 111

4.2.2 Bias ... 111

4.2.3 Confounding ... 112

4.2.4 Causality ... 113

5. Outlook ... 119

6. References ... 125

(9)

List of Figures

Figure 1: Clinical trials are carried out in a step-wise manner. Each phase of the trial aims to

answer a specific question. ... 2

Figure 2: Settings of case-control and cohort-studies. ... 9

List of Tables Table 1: Bradford Hill criteria and their respective meanings used to assess causality in pharmacoepidemiologic studies. ... 6

Table 2: Differences between cohort- and case-control studies, and advantages and limitations of both study designs. ... 11

Table 3: Overview of the different data types available to researchers from the CPRD ... 15

Table 4: Causes of seizures overall and by age group. ... 18

Table 5: Additional risk factors for seizures ... 18

Table 6: Classification and key features of focal and generalized seizures. ... 20

Table 7: Characteristics and lifestyle factors of cases with seizures and matched controls (project 3.1) ... 36

Table 8: Odds ratios of psychiatric comorbidities among cases with seizures and matched controls (project 3.1) ... 37

Table 9: Odds ratios of psychiatric comorbidities among cases with seizures and matched controls,by timing of the diagnosis (project 3.1) ... 38

Table 10: Odds ratios of neurologic comorbidities among cases with seizures and matched controls (project 3.1) ... 39

Table 11: Odds ratios of neurological comorbidities among cases with seizures and matched controls,by timing ofthe diagnosis (project 3.1) ... 40

Table 12: Odds ratios of selected drug groups among cases with seizures and matched controls,by current or past use, and by current use and number of prescriptions (project 3.1)41 Table 13: Incidence rates of seizures in patients with depression with no antidepressant treatment, with current and past use of antidepressants, and with current use of different antidepressant drug classes, by age or sex (project 3.2) ... 55

Table 14: Incidence rates of seizures in current users of most frequently used single antidepressants, by age or sex (project 3.2) ... 57

Table 15: Characteristics of cases with seizures and matched controls (project 3.2) ... 59

(10)

Table 16: Odds ratios for seizures in users of different antidepressant drug classes, by current or past use, by current use and sex, or by current use and number of prescriptions prior to the index date (project 3.2) ... 61 Table 17: Odds ratios for seizures in users of different antidepressant single drugs, by use and by dose used at the index date (project 3.2) ... 63 Table 18: Odds ratios for seizures in users of different antidepressant drug classes compared to non-users of antidepressants, by current or past use, and by dementia (project 3.2) ... 65 Table 19: Odds ratios for seizures in users of different antidepressant drug classes compared to non-users of antidepressants, by current or past use, and by a history of stroke/TIA (project 3.2) ... 67 Table 20: Odds ratios for seizures in users of different antidepressant drug classes compared to non-users of antidepressants, by current or past use (project 3.2) ... 69 Table 21: Odds ratios for seizures in users of different antidepressant drug classes compared to non-users of antidepressants, by current or past use, and by switching of antidepressants during follow-up, compared to non-users of antidepressants (project 3.2) ... 70 Table 22: Incidence rates of seizures in patients with no antipsychotic treatment, with current and past use of antipsychotics, and with current use of different antipsychotic subclasses or single drugs, by underlying disorder (project 3.3) ... 86 Table 23: Characteristics of cases with seizures and matched controls,by underlying disorder (project 3.3) ... 87 Table 24: Odds ratios for seizures in users of different antipsychotic subclasses, by current or past use, and by underlying psychiatric disorder (project 3.3) ... 90 Table 25: Odds ratios for seizures in users of different antipsychotic single substances, by current or past use, and by underlying disorder (project 3.3) ... 92 Table 26: Odds ratios for seizures in users of different antipsychotic drug groups, by current or past use, and by history of stroke or TIA prior to the index date (project 3.3) ... 94 Table 27: Odds ratios for seizures in users of different antipsychotic drug groups, by current or past use and age, and by current use and number of prescriptions prior to the index date, and by underlying disorder; reference group non-users of the respective drug group (project 3.3) ... 96 Table 28: Discussion of the criteria used to assess the likelihood of causality with regard to the associations observed in projects 3.1-3.3 of the thesis ... 114

(11)

Summary

i

I. Summary

Pharmacoepidemiology studies use and effects of drugs in large numbers of people. It allows the investigation and quantification of rare beneficial or adverse events of drugs used by the general population under “real-world” conditions. Pharmacoepidemiologic research strongly depends on and has been facilitated by the development of large scale health care databases.

Among these the U.K. Clinical Practice Research Datalink (CPRD) stands as one of the largest and best validated medical records databases worldwide. The CPRD was initiated more than 25 years ago and contains records on diagnoses, drug prescriptions, demographics, lifestyle variables and medical procedures performed from over 12 million patients contributing 64 million person-years of prospectively recorded primary healthcare data.

CPRD data was employed in all studies carried out in this thesis. The goal was set to identify and analyze risk factors for new-onset seizures in patients with neuropsychiatric disorders.

While it has long been suspected that patients suffering from neuropsychiatric disorders exhibit an increased risk of new-onset seizures no significant real-world evidence exists on risk factors associated with these seizures.

We first investigated risk factors for new-onset seizures in adult patients with depression. Our results suggest that patients suffering from depression were at an increased risk of seizures if they abused drugs, suffered from alcoholism, had a history of cerebrovascular disease or recent brain injury, comorbid dementia, or comorbid psychiatric disorders. Additionally we found current users of cephalosporins or antiarrhythmics to be at an increased risk of seizures compared with non-users of these drug classes.

In a follow-up study we assessed the association between antidepressant drug use and new- onset seizures in adult patients with depression. Our data suggest that the absolute risk of seizures in this population was rare, irrespective of whether patients used antidepressants or not. Additionally we found that the use of selective serotonin reuptake inhibitors (SSRIs) or serotonin norepinephrine reuptake inhibitors (SNRIs) was associated with a twofold increased risk of seizures compared to non-use. However, tricyclic antidepressants (TCAs) at low doses, as prescribed in this primary care setting, were not associated with seizures. Among users of SSRIs and SNRIs, treatment initiation was associated with a higher risk of seizures compared to longer-term treatment. Finally, we could demonstrate that higher doses of antidepressants

(12)

Summary

ii

prescribed were associated with an increased risk of seizures than lower doses, although small sample sizes limited conclusiveness.

In the final study of this thesis, the potential association between antipsychotic drug use and new-onset seizures among patients with different underlying neuropsychiatric disorders was investigated. The results obtained in this study demonstrate that the association between antipsychotic drug use and seizures was strongly modified by the underlying neuropsychiatric indication. Our data shows that patients with dementia exhibited a significantly higher risk of seizures than patients with affective disorders, irrespective of the use of antipsychotics.

Additionally, in patients with affective disorders, current use of medium to high potency first- generation antipsychotics (haloperidol, prochlorperazine, or trifluoperazine) was associated with a more than twofold increased risk of seizures compared to non-use of antipsychotics. In all of these patients, use of all other antipsychotics was not associated with new-onset seizures. In patients with dementia, current use of the second-generation antipsychotics amisulpride, aripiprazole, risperidone, or sulpiride, was not associated with seizures, while current use of all other antipsychotics was associated with an increased risk of seizures.

We found that the inability to adjust for confounding by disease severity, the unproven validity of the diagnoses of affective disorders and seizures, and the limited sample sizes in sub-analyses posed a certain limit to our studies.

Nevertheless, all studies carried out in this thesis provide new insight into the poorly understood relationship between neuropsychiatric disorders and new-onset seizures. Formally quantifying the occurrence of seizures and assessing risk factors for seizures among this restricted study population was only feasible through access to the large existing data set comprising detailed patient information available from the CPRD.

(13)

Abbreviations

iii

II. Abbreviations

ADHD Attention deficit hyperactivity disorder

AMPA Alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid BCE Before the Christian Era

BCDSP Boston Collaborative Drug Surveillance Program

BMI Body mass index

BPU Basel Pharmacoepidemiology Unit

CI Confidence interval

CNS Central nervous system

CPRD Clinical Practice Research Datalink

CT Computed tomography

DDD Defined daily dose

EEG Electroencephalogram

FFDCA Federal Food, Drug and Cosmetic Act

GP General Practitioner

GABA Gamma-aminobutyric acid

GPRD General Practice Research Database

IR Incidence rate

MRI Magnet resonance imaging

NA Not applicable

NMDA N-Methyl-D-aspartic acid

OR Odds ratio

PYs Person-years

RCT Randomized Controlled Trial

SNRI Serotonin norepinephrine reuptake inhibitor SSRI Selective serotonin reuptake inhibitor TCA Tricyclic antidepressant

TIA Transient ischemic attack

U.S. United States

U.K. United Kingdom

VAMP Value Added Medical Products

(14)

(15)

Introduction Pharmacoepidemiology

1

1. Introduction

1.1 Pharmacoepidemiology

1.1.1 Defining pharmacoepidemiology

Pharmacoepidemiology, a subdiscipline of epidemiology, investigates use and effects of drugs in large numbers of people.^1,2 Combining clinical pharmacology (i.e., the study of effects of drugs in humans) with epidemiology (i.e., the study of factors that influence or determine the occurrence and distribution of health-related states or events in populations),^2–4 pharmacoepidemiology is primarily used to study large populations of individuals, contrary to clinical pharmacology.¹

1.1.2 Historical development of pharmacoepidemiologic studies

Before the 1950s, little proof was demanded in terms of safety and efficacy prior to introducing new drugs to the market.¹ Only upon occurrence of harmful events associated with drug use, regulatory actions were taken to ensure drug safety.

One of the most far-reaching drug scandals was the extensive prescription of thalidomide, a sedative drug, to pregnant women.⁵ At the time thalidomide showed no toxic effects in animals and was thus assessed and advertised as safe for use in pregnancy.⁵ Thalidomide was introduced to the European market in 1957, and withdrawn only four years later.⁵ During this period of time, it caused birth defects in more than 10,000 children who were born with missing limbs or limb anomalies.⁵

In 1962, the U.S. government thus made amendments to the Federal Food, Drug and Cosmetic Act (FFDCA), requiring formal proof of efficacy and relative safety in terms of risk-to-benefit ratio for any disease to be treated.⁶ The 1962 amendments to the FFDCA enforced the process leading to the establishment of so-called clinical trials, which are nowadays standard procedure prior to placing a new drug on the market.⁶

(16)

2

1.1.3 Clinical trials and the role of pharmacoepidemiology

Clinical trials are used to investigate pharmacokinetic and pharmacodynamic properties of a drug, and to evaluate a drug’s efficacy, safety, and tolerability.⁶ They are carried out in a series of subsequent steps, so-called phases, where each phase is designed to investigate a specific aspect of the process of drug development (see Figure 1 for details).

Figure 1: Clinical trials are carried out in a step-wise manner. Each phase of the trial aims to answer a specific question. Figure adapted from⁶.

1.1.3.1 Clinical trials: phase I to III

Phases I to III of clinical trials are limited in sample size and observational duration (as shown in Figure 1).

These initial phases of clinical trials are conducted in a well-defined, yet artificial environment, where specific patient subgroups (e.g., elderly patients, polymorbid patients who use various drugs, women of childbearing age, and children) are excluded from the analysis.⁶

(17)

3

1.1.3.2 Clinical trials: phase IV

As the drug is placed on the marked and thus used by the general population, including all patient subgroups, phase IV of clinical trials is initiated. The so-called post-marketing surveillance phase studies beneficial and adverse effects of drugs in the general population.³ Since drugs are often used over longer periods of time in “real life” than in phases I to III of clinical trials, phase IV investigates rare or delayed adverse or beneficial effects that were not noticed prior to market introduction.¹

The post-marketing surveillance comprises pharmacovigilance and pharmacoepidemiology studies. Pharmacovigilance studies are based on spontaneous reporting systems of adverse drug events, and are important to detect signs of adverse events not seen in phases I-III of clinical trials.⁷ Such reports are however difficult to interpret; only a small (unknown) proportion of suspected adverse events of drugs are reported spontaneously, and adverse events are more likely reported if they are serious, or if the drug has received a lot of media attention.^3,7,8

Formal quantification and investigation of adverse drug events is done in pharmacoepidemiologic research.^1,8 Therefore, it is essential to know the number of people exposed and not exposed to the drug under investigation, as well as the number of people in each group who developed the outcome in question. In conclusion, the major goal of pharmacoepidemiologic research is to describe drug utilization patterns under

“real life” conditions and to quantitatively investigate previously undetected beneficial or adverse drug effects.¹

1.1.4 Association types in pharmacoepidemiology

Pharmacoepidemiology investigates potential associations between drug exposure and beneficial or adverse outcomes. An association is defined as two events (i.e., drug exposure and outcome under investigation) occurring together repeatedly, with this repeated occurrence taking place more often than a chance occurrence.³ The following four scenarios can result from investigating such potential associations: No association, artifactual associations, indirect associations, or causal associations.⁹

(18)

4

1.1.4.1 No association

No association between exposure and outcome is observed when exposure and outcome are in fact independent of one another or when the power of the study was too low to detect an association.

The power of a study can be defined as the probability to detect an association between exposure and outcome if the association exists.² The power increases with the precision of the outcome variable, the strength of the association, the sample size, and increased α- level (refer to 1.1.4.2).¹⁰

1.1.4.2 Chance and bias – two types of artifactual associations

Chance (i.e., random error), is a random variation due to every study being performed on only a sample of the entire population while inferring from this subset to the whole population. Depending on the analyzed sample results will vary due to irregular variations. Random errors associated with the results obtained from analyzing such subsets will decrease with increasing sample size.^2,4,8

Whether associations observed between exposures and outcomes are due to chance or not can be assessed by statistical testing. First, an (arbitrarily) selected level, the α-level, is set. The α-level specifies the probability P that an association is due to chance only when in reality no association exists.⁴ As a standard, an α-level of 0.05 is chosen, referring to the P-level reported in most pharmacoepidemiologic studies.^4,10A reported association between a specific exposure and outcome with a P-value smaller than 0.05 (P<0.05) is interpreted as a statistically significant association, thus ruling out the possibility of a chance finding with a confidence of greater or equal to 0.95 (≥ 95%).^8,9 P-values do not convey a message on the strength or the relevance of an association; they simply convey a message on how unlikely an observed association would be if in reality there was no association.¹⁰

In contrast to chance, bias (i.e., systematic error), is a systematic variation due to treating or evaluating two study groups consistently differently. Unlike chance, bias is insensitive to sample size and cannot be ruled out by statistical testing.⁸

(19)

5

Bias can cause an apparent association between exposure and an outcome when in reality no association exists.⁹ It is therefore important to prevent bias by selecting a proper study design since it cannot be controlled for after study completion. The literature distinguishes broadly between two types of bias: selection and information bias.

Selection bias refers to situations in which the estimated effect of an exposure on an outcome is distorted due to procedures used to select subjects for study participation.^2,8 For example, if study participation is voluntary, participators might differ systematically in exposure and outcome status from those declining participation; since the association between exposure and outcome among nonparticipants is unknown, so is the degree of the systematic error introduced.⁸

Information bias refers to situations in which the estimated effect of an exposure on an outcome is distorted because the information collected on exposure and/or outcome is erroneous.⁸ Common types of information bias are measurement bias (i.e., measurement of exposure or outcome is not done in a comparable way in groups to be compared), recall bias (i.e., individuals affected by the outcome remember exposures differently than individuals not affected by the outcome), or interviewer bias (i.e., interviewers behave differently with the groups to be compared).⁸

1.1.4.3 Indirect associations

Indirect (“spurious”) associations result from confounding. Confounding variables are variables other than the exposure or outcome under investigation. They are associated with the exposure (without being an effect of the exposure) and are risk factors for the outcome, creating an apparent association or masking a real association between exposure and outcome.^8,9

An example of a confounder is the apparent association between wearing leather shoes in bed at night and suffering from a headache the morning after. One could claim that wearing leather shoes at night causes headaches, when in reality it is heavy drinking associated with forgetting to take off the shoes before going to bed.¹¹

Confounding in pharmacoepidemiologic studies can be limited if the existence of a confounder is known and if the confounder can be measured.⁸ This can be done by

(20)

6

restricting the study population (i.e., exclusion of individuals with potential confounders), matching cases and controls (in case-control analyses) or exposed and non-exposed individuals (in cohort studies) on potential confounders, stratifying the analyses by the potential confounder, or mathematical adjustments in the analysis.⁸

1.1.4.4 Causal associations

The primary aim of pharmacoepidemiologic research is to determine whether a certain exposure is causally associated with a specific outcome. However, statistical testing cannot assess causality and thus an observed association can, even when statistically significant, be simply due to bias or confounding (see previous chapters for a discussion of these factors).

Nevertheless, certain criteria that were developed in 1965 by Sir Austin Bradford Hill,¹² can be applied to assess the likelihood of a causal association. The most important of these criteria for pharmacoepidemiologic research, and their respective meanings, are summarized in Table 1.

Table 1: Bradford Hill criteria and their respective meanings used to assess causality in pharmacoepidemiologic studies (summarized from¹²).

Criterion Meaning

Strength of the association The stronger the association the more likely a causal link.

Consistency The association has been observed by different investigators, in different places and times, and under different circumstances.

Temporality The exposure must precede the outcome.

A typical pitfall: a drug is taken for early signs of a disease which has not yet been diagnosed. The temporal sequence suggests that the drug causes the disease, when in reality the disease started before the drug exposure.

Dose-response relationship The higher the exposure (e.g. an increase in drug dose) the greater the effect (e.g.

adverse event).

Biological plausibility/Coherence with existing information

Good general theory to explain a causal link.

The association makes sense taking into account the evidence from the literature;

however this criterion strongly depends on the available literature.

This criterion is never absolute, as knowledge changes over time, and so does plausibility that something is true.

Experimental evidence Changing the exposure under controlled conditions causes a change in the outcome.

Reversibility Removal of exposure leads to decline in outcome.

(21)

7

None of these “Bradford Hill criteria” are considered sufficient to prove causality.¹² However, the more criteria are met, the more likely an association is causal.⁹ Thus, any observed association in pharmacoepidemiology has to be interpreted in the context of the best evidence available at the time the study is carried out.⁹

1.1.5 Study designs in pharmacoepidemiology

Pharmacoepidemiologic research is frequently categorized into observational studies (descriptive studies, analytical studies) and interventional studies (e.g. randomized controlled trials).¹³ Depending on the purpose of the research, some study designs are better suited than others (refer to the subsequent chapters for discussion of the different study types).

1.1.5.1 Observational studies

Observational studies are categorized into descriptive or analytical studies and are used to investigate drug utilization patterns and drug safety.¹³ A major advantage of observational studies is that they can be applied if interventional studies are unethical (e.g., if an exposure is known to be harmful), unnecessary (e.g., if an intervention is already proven to be efficient), or not feasible (e.g., if the outcome is rare or delayed).¹⁴

1.1.5.1.1 Descriptive studies

Descriptive studies are employed to characterize existing distributions of exposures or outcomes without investigating causal inferences.² Descriptive studies are useful for generating hypotheses, although they cannot be applied to determine whether exposure or outcome occurred first.¹⁵ Additionally, due to the lack of comparison groups, no causal associations can be determined.^2,13,15 In the following, the different types of descriptive studies including case reports, case series, ecologic studies, and cross-sectional studies are discussed.

Case reports describe the experience of one patient while case series describe the experience of several patients when using a particular drug (e.g., what clinical features are observed after a drug overdose?). Case reports and case series do not provide sufficient evidence for making causal inferences, but they often give rise to hypotheses.¹³ For

(22)

8

example, several authors reported seizures as an adverse event in patients using second- generation antipsychotics such as quetiapine, aripiprazole, or olanzapine, in therapeutic or in overdoses.^16–20

Ecologic studies do not investigate data from individual patients, but from groups. These studies can be useful to describe differences in the prevalence of certain exposures and outcomes between groups or countries.¹³ However, no conclusions can be drawn at the individual level, because confounding factors of the individuals are unknown.¹³ In a recent study the correlation between antidepressant prescribing patterns in Organization for Economic Co-operation and development (OECD) countries and the rates of suicide was assessed.²¹ While a weak positive correlation between antidepressant use and suicide was observed between countries, no inference could be made at a person’s individual risk of suicide associated with the use of antidepressants.²¹

In cross-sectional (prevalence) studies, a certain exposure or outcome in a population is assessed at a specific point in time, or during a specific time span. The prevalence of an exposure or outcome is defined as the total number of individuals who have the exposure or outcome under investigation at a particular time (or over a particular time span), divided by the population at risk of having this exposure or outcome at that time (or over that particular time span).² The prevalence of an exposure or outcome in a population is often reported as a percentage. Examples of cross-sectional studies are investigations on lifetime prevalence estimates of major depression across different countries.^22,23

1.1.5.1.2 Analytical studies

Analytical studies assess and quantify the association between exposures and outcomes.² Cohort and case-control studies are two main study types used to test hypotheses on the etiology or risk factors for an outcome.

Cohort studies identify the study population based on exposure status, as described in the following two paragraphs summarized from²⁴. The population is followed over time to investigate differences in outcomes between the different exposure groups. In their simplest form, cohort studies compare exposed individuals (i.e., individuals with the alleged risk or protective factor) to unexposed individuals (i.e., individuals not having this

(23)

9

factor) with regard to subsequent outcome frequency. In more elaborate settings, different exposure groups can be studied simultaneously.

In cohort-studies, the study population is initially outcome-free and the occurrence of the outcome is measured over time. This type of study can be performed prospectively (i.e., data on exposure and outcomes is collected while the study is conducted) or retrospectively (i.e., data on exposure and outcomes is already available at the time the study is conducted) (see Figure 2, upper part).

Figure 2: Settings of cohort and case-control studies (adapted from^24,25).

For example, a British cohort study published in 2011 assessed the association between the use of tricyclic and related antidepressants, monoamine oxidase inhibitors, selective serotonin reuptake inhibitors, or other antidepressants, and severe adverse events, in depressed patients aged 65 or over.²⁶

(24)

10

In a case-control study, the study population is identified based on outcome status. Cases with an outcome are compared with controls without the outcome, looking for differences in antecedent exposures (see Figure 2, lower part).²⁵

In a case-control study conducted between 1995 and 1999, the cases were all Icelandic patients aged 10 years or over who had a first unprovoked seizure; for each case, a suitable age-matched control who did not have seizures up to the time the case had a seizure was identified.²⁷ The odds ratio of antecedent major depressive disorder was 1.7- fold increased among cases compared with controls, suggesting that major depressive disorder is a risk factor for first unprovoked seizures.²⁷

Table 2 summarizes advantages and limitations of cohort studies and case-control studies.

(25)

11

Table 2: Differences between cohort- and case-control studies, and advantages and limitations of both study designs.9,13,24,25,28

Cohort study Case-control study Comments

Sample size Large Usually smaller than cohort study

Time Slow Rapid Not an issue for retrospective database

research using medical records

Cost High Usually smaller Not an issue for retrospective database

research using medical records Analysis Computationally more complex than case-control studies Computationally easier than cohort studies Especially if time-dependent exposures

are analyzed More suited for Rare exposures: special groups with a high frequency of

the exposure can be identified and included in the study population

Rare outcomes: a sufficiently large number of cases with the outcome can be included

Outcomes to be studied Multiple Just one In case-control studies, it is important to

ensure that the exposure occurred prior to the outcome

Exposures to be studied Just one Multiple

Measures All measures, including incidence rates, attributable risks, and relative risks

Only odds ratio: incidence rates cannot be calculated as the size of the population at risk is unknown

The odds ratio is a valid estimate of the relative risk if: 1) the cases are

representative of the population at risk, 2) the controls are randomly selected from the population giving rise to the cases, 3) the outcome is rare in the population at risk (<5%)

Validity of exposure data Generally higher than in case-control studies (as participants are selected based on their exposure)

Generally lower than cohort studies: retrospective assessment of exposure limits its validity

Not an issue for retrospective database research using medical records

(26)

12 Table 2 (cont.)

Cohort study Case-control study Comments

Comparison of exposed and unexposed groups/ cases and controls

Exposed and unexposed individuals must be as similar as possible at baseline in all aspects except the exposure under investigation

Controls should be selected from the same population that gave rise to the cases

Causal relationships More reliable: all subjects are outcome-free at the beginning (temporal relationship given), and exposure information can be more accurately assessed

Less reliable: temporal relationship not as easily identified, exposure information can be biased

Not an issue for retrospective database research using medical records

(27)

13

Generally one refers to a nested case-control study when the population within which a case- control study is conducted is well defined.⁸ Nested-control studies rely on the advantages of both cohort and case-control studies.

The following paragraph represents a summary of²⁸:

In a nested case-control study a cohort of individuals is followed until they develop the outcome under investigation or until their follow-up ends due to other reasons (e.g., death or loss to follow-up). The analysis is then conducted as a case-control analysis. The individuals who developed the outcome are defined as cases, and the date of their outcome is named index date. The individuals who did not develop the outcome up to the index date of the case are potential controls. Instead of analyzing the whole original cohort of individuals (as is done in cohort studies), only the cases and a defined number of all potential controls are analyzed. Thus, nested case-control studies are well suited to investigate time-dependent exposures (e.g., drugs in a large cohort of individuals followed over many years). In such cases, nested case-control analyses are computationally less complex than analyses of the entire cohort.

Additionally, nested case-control studies are advantageous if exposure information is costly (e.g., analyses of blood samples), and when seeking this information for everyone in the cohort is too expensive.⁸

1.1.5.2 Experimental (interventional) studies The following paragraph represents a summary of⁴:

Experimental studies are cohort studies in which the exposure status is determined by the investigators. This study design allows testing the efficacy or risk of an intervention. In randomized control trials (RCTs), the intervention is allocated randomly (i.e. the assignment is unpredictable). The major strength of this design is that investigators cannot allocate the intervention, thus all patients are as likely to receive it. Thus, in large RCTs the groups with or without the intervention are likely to be similar with regard to potential confounding variables (e.g., age, sex, severity of disease).

As selection bias and confounding are avoided, associations demonstrated in RCTs are more likely to be causal than those demonstrated in observational studies.⁹ However, RCTs lack

(28)

14

generalizability, since they cannot assess safety issues in large populations of individuals who use drugs under everyday conditions in clinical practice.²⁹

1.1.6 Data sources in pharmacoepidemiology

The increased availability of large health care databases has facilitated pharmacoepidemiologic studies. Thus, drug utilization or safety studies can be conducted in subgroups of patients that were excluded from clinical trials. Additionally, if multiple years of patient follow-up data are available, assessment of long-term safety of drugs is possible.³ Available databases can be divided into the two main categories medical records databases (data is collected by physicians [usually primary care providers] who enter information on patients while providing medical care) and administrative databases (data is collected primarily for administrative purposes, such as reimbursement of health care services).^29,30 One of the largest and best validated medical care databases worldwide is the U.K. Clinical Practice Research Datalink (CPRD).³¹ This database was established in 1987 as ‘Value Added Medical Products (VAMP)’ database, and was long known as ‘General Practice Research Database (GPRD)’.³² In 2011, the CPRD contained records from more than 12 million patients contributing 64 million person-years of prospectively recorded primary healthcare data.³³

The U.K. health care system is particularly suitable for establishing primary care research databases such as the CPRD, as almost every person in the U.K. is registered with a general practitioner (GP) who functions as a ‘gate-keeper’ in the healthcare system. Secondary care (specialist care, hospital care, etc.) is either provided at the request of the GP, or if directly assessed, full disclosure from secondary care is reported back to the GP.³³ Through this prospective recording by the GPs and the nearly complete medical histories of patients, this database is especially valuable for pharmacoepidemiology.

While prescription data is nearly completely documented as the GPs use the computer to generate prescriptions, diagnoses must be manually recorded and can therefore be incomplete or wrongly coded.³⁴ Although the validity of the diagnoses recorded in the CPRD is high (with an average positive predictive value of almost 90%), records of acute disorders may be less valid than records of chronic disorders.^34,35

(29)

Introduction

15

Today, researchers can access through a linkage system data from secondary care hospital episode statistics, death certification data, socioeconomic classification data, and disease registry data including the National Cancer Intelligence Network and the Myocardial Infarction National Audit Program register.³³

Table 3 summarizes different data types available to researchers from the CPRD.

Table 3: Overview of the different data types available to researchers from the CPRD.³²

Data Details

Patient characteristics Gender, birth year, weight, height, body mass index, occupation, ethnicity, marital status etc.

Patient demographics Practice attended, U.K. region the patient lives in

Medical diagnoses Recorded as “Read codes”

Drug prescriptions Including quantity of the prescribed drug, and dose instructions Lifestyle variables Smoking status, alcohol consumption

Referrals to hospitals or specialists With diagnoses, drug prescriptions

Laboratory tests E.g. blood and urine tests

(30)

Introduction Seizures and Epilepsy

16

1.2 Seizures and Epilepsy

1.2.1 Defining seizures and epilepsy

Seizures are transient disruptions of brain function resulting from excessive neuronal activity.³⁶ Depending on location and extent of the affected brain regions, seizures result in alterations of muscle tone, sensations, consciousness, or behavior.³⁶ Seizures are clinical events that can occur in all people for example after sleep withdrawal, overdoses or withdrawal from certain drugs, or hypoxia.³⁷

Epilepsy is a chronic condition of repeated spontaneous seizures.³⁶ Spontaneous seizures are defined as seizures without a direct precipitating cause. A diagnosis of epilepsy requires either at least two spontaneous seizures occurring more than 24h apart, one spontaneous seizure with a high risk of recurrence, or the diagnosis of an epilepsy syndrome (defined by a cluster of symptoms occurring together including seizure type, etiology, age of onset, and other factors).^38,39 Epilepsy is a disorder of the brain that is characterized by an enduring predisposition to generate seizures.³⁷

1.2.2 History of seizures and epilepsy

Seizures and epilepsy have been studied by medical scholars throughout time. One of the earliest publication on epilepsy, “the Sacred Disease”, part of the Hippocratic Collection, was written in 400 BCE, potentially by Hippocrates himself.⁴⁰ It was suggested that epilepsy is a hereditary disorder of the brain, and thus contemporary beliefs that seizures and epilepsy were a sign from the gods were refused⁴⁰:

“My own view is that those who first attributed a sacred character to this malady [i.e., epilepsy] were like the magicians, purifiers, charlatans and quacks of our own day, men who claim great piety and superior knowledge. Being at a loss, and having no treatment which would help, they concealed and sheltered themselves behind superstition, and called this illness sacred, in order that their utter ignorance might not be manifest.”

While this early report showed profound knowledge of the human body, it was not until very recently that the findings of Hippocrates were acknowledged. For example, throughout the Middle Ages and the Renaissance, patients with epilepsy were thought to be possessed by the devil and women with epilepsy were persecuted as witches.⁴¹

(31)

17

It was not until the mid-19^th century, when John Hughlings Jackson recognized that seizures were caused by “occasional, sudden, excessive discharges of gray matter”.³⁹ Jackson discovered that seizures could spread from a single focus with localized motor symptoms to generalized seizures accompanied by loss of consciousness.^36,39

Following these findings, bromide, the first antiepileptic drug, became available, followed by phenobarbital (1912) and phenytoin (1937).^36,39 With the development of new techniques, such as the electroencephalogram (EEG) in 1929, neuroscientists were able to show that seizures are associated with neuronal hyperexcitability in the brain.³⁹

1.2.3 Epidemiology of seizures and epilepsy

Worldwide, it has been estimated that up to 10% of people have at least one seizure at some point in their life, and 0.4 - 1% of people suffer from epilepsy.^42,43

In developed countries the overall incidence rate of spontaneous seizures was estimated 55 per 100,000 person-years.^44,45 These data are not conclusive with regard to potential gender differences in seizure risk.^44–46 Spontaneous seizures occur most frequently in neonates and infants (incidence rate 100 to 130 per 100,000 person-years) and people aged 65 years or over (incidence rate 110 to 180 per 100,000 person-years).^44,46

Focal seizures with or without generalization are assumed to be more common than generalized-onset seizures; however, in most studies misclassification of seizure types is difficult to estimate (see chapter 1.2.6).^44,45,47

Estimates of the overall incidence rate of epilepsy range from 30 to 50 per 100,000 person- years in high income countries to 120 per 100,000 person-years in low-income countries.^42,43,47 The higher incidence of epilepsy in low-income countries compared to high- income countries is presumably caused by parasitic diseases associated with seizures, such as malaria or neurocysticercosis, as well as lower standards in medical infrastructure.⁴⁷ It was shown that incidence rates of epilepsy exhibit similar age-related trends as incidence rates of spontaneous seizures, with peaks in the first year of life and at ages 65 or over.^44,45

Focal epilepsies, commonly caused by localized tumors, developmental malformations, or damages after head trauma or stroke, account for about 60% of all epilepsies. Generalized epilepsies, mostly based on genetic mutations, account for the remaining 40% of epilepsies.³⁹

(32)

18

1.2.4 Risk factors for seizures and epilepsy

Risk factors for seizures and epilepsy vary by age groups, as Table 4 displays. Across all age groups however, the causes of seizures remain unknown in most cases.

Table 4: Causes of seizures overall (%) and by age group (: frequent cause in this age group).

Causes

Age group

Unidentified cause44 Cerebrovascular disease44 Neoplasms 44 Traumatic brain injury46,48,49 Cerebral palsy/intellectual disability 44 Infection46,48,49 Neurodegenerative diseases 44,46 Other 46,48,49

All ages 68% 9% 6% 5% 4% 1% 7% 1%

Neonates and children    

Middle aged    

Adults aged 65   

Numerous additional risk factors for seizures have been identified (see Table 5).

Table 5: Additional risk factors for seizures

Category Risk Factor References

Additional risk factors for seizures

Heavy alcohol consumption or alcohol withdrawal ^50–53

Illicit drug use ⁵⁴

Medication use or withdrawal ^55–60

Metabolic or electrolyte imbalances ⁶¹

Fever ⁶¹

Severe dehydration ⁶²

Sleep deprivation ^63,64

Anoxic encephalopathy ⁶⁰

Additional disorders associated with an increased risk of seizures

Depression and suicidality ^27,65,66

Other psychiatric disorders ^66,67

Migraine with aura ^68,69

Severe hypertension ^70,71

Attention deficit (hyperactivity) disorder ⁷²

Multiple sclerosis ^73,74

Systemic lupus erythematosus ⁷⁴

Preeclampsia ⁶⁰

(33)

19

1.2.5 Mechanism of seizures

Little is known about the physiological processes underlying seizures. The following paragraphs summarize the current knowledge about how seizures may be generated by the brain^36,39,75:

Glutamate is the major excitatory transmitter while GABA (-Aminobutyric acid) is the main inhibitory transmitter in the brain. It is currently believed that excessive synaptic activity during seizure results from an imbalance between these two antagonistic neurotransmitters.

Depolarization and thus neural excitation is triggered by two types of glutamate regulated channels, AMPA (-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) and NMDA (N- Methyl-D-aspartic acid), in combination with voltage gated sodium and calcium channels.

Hyperpolarization and therefore neural inhibition is caused by activation of GABA receptor- mediated chloridechannels and different types of potassiumchannels.

Under normal conditions, post-hyperpolarization following an action potential prevents immediate generation of a new action potential and thus neuron hyperexcitation. However, during a seizure, neurons located in the seizure focus exhibit a prolonged depolarization phase (depolarization shift), which can be identified as a sharp waveform in the EEG of patients. The observed depolarization shift is followed by rapid firing of action potentials in affected neurons. During such an event GABAergic inhibition of regulatory signals appears to be suppressed, enabling the spread of neural hyperexcitability to surrounding areas in the brain.

1.2.6 Seizure types

Focal seizures start within networks of one brain hemisphere, and are classified based on whether they affect consciousness or awareness.⁷⁶ Focal seizures can evolve to generalized seizures affecting both hemispheres.⁷⁶

Generalized seizures begin at some point within, and rapidly engage networks distributed in both brain hemispheres.⁷⁶

Classification and features of focal and generalized seizures are summarized in Table 6.^43,76–78

(34)

20

Table 6: Classification and key features of focal and generalized seizures. Summarized from^43,76–78.

Typical manifestations Consciousness Duration

Focal seizures Without impairment of

consciousness or awareness

Motor symptoms: jerking, spasms, posturing, reversible weakness in one side of the body (Todd’s paralysis)

Sensory- or psychic symptoms (i.e. aura): tingling, numbness, pain, feeling of heat,

hallucinations (flashing lights), sudden intense emotions (fear, depression, anger, irritability), dysphasia, disturbance of memory, sensations of unreality or depersonalization, hallucination of vision, taste, or smell

Preserved Mostly only a few seconds

With impairment of consciousness or awareness

Altered consciousness (confusion, unresponsiveness, motor arrest)

Automatisms (repetitive movements such as chewing, lip smacking, fiddling, tapping, whistling, humming, uncoordinated violent behavior)

Impaired or lost Few seconds to minutes

Generalized seizures

Tonic Stiffening of muscles (e.g. contraction of facial or respiration muscles, rising up of arms) Falling over, usually backward

Impaired or lost

Less than 60 s, person recovers quickly

Clonic Jerking or twitching of limbs or body About 2 min

Myoclonic Brief contraction of muscles, resulting in sudden irregular jerking or twitching of trunk or one or more limbs

Sometimes caused by drugs (antidepressants, antipsychotics), drug toxicity or withdrawal, post-hypoxic brain damage

Presentation varies from subtle jerk to violent jolt

Fraction of a second

Atonic Sudden relaxation of muscles

Person often falls over, usually forward (also called “drop attacks”)

Less than 60 s, person recovers quickly Tonic-clonic (“grand mal”) Loss of consciousness, tonic phase with flexion and rigidity

Clonic phase with convulsions of usually all four limbs

Stertorous breathing, froth of saliva from mouth, loss of bladder control, cyanosis

Tonic phase: 10-30 s Clonic phase: 30-60 s Confusion after: > 10 min Absence (“petit mal”) Abrupt loss of consciousness and cessation of motor activity, staring vacantly into space

Mostly in children or adolescents, continue into adulthood in 7-80% of cases EEG diagnostic in more than 90% of cases

Mostly less than 10 s

(35)

21

1.2.7 Diagnosis of seizures

Diagnosis of an epileptic seizure mainly relies on the patient’s clinical history, rendering eye witness reports of seizures crucial in establishing a correct diagnosis.⁷⁸

Although no features exclusively correlated with epileptic seizures are known, a handful of strong seizure markers have been identified to this point: postictal confusion,^78,79 occurrence out of sleep,⁷⁹ cyanosis,^78,79 lateral tongue biting,^78,79 preceding “déjà vu” or “jamais vu”,^78,79 confirmed unresponsiveness, head or eye turning to one side,^78,79 rhythmic limb shaking,⁷⁸ and unusual posturing.^78,79

Following an initial seizure event, an EEG recording is done to detect potential abnormalities in electrical activity. However, less than half of patients undergoing epileptic seizures show detectable EEG abnormalities within 24h after seizure occurrence.⁸⁰ Additionally, magnet resonance imaging (MRI) or computed tomography (CT) are used to detect direct causes of seizures, such as brain injuries, brain tumors, or stroke.⁶¹

A variety of medical conditions imitate symptoms of epileptic seizures leading to one third of false epileptic seizure diagnoses.⁷⁸ For example, about 20% of patients referred to epilepsy centers have psychogenic nonepileptic seizures, by definition a psychiatric, not a neurologic disorder.⁷⁹ This type of seizures is characterized by a resistance to antiepileptic drugs, unusual triggers, a tendency to occur in the presence of an audience, a history of psychiatric diagnoses, and the presence of a normal EEG during video monitoring of the seizure.⁷⁹

Differential diagnosis of epileptic seizures should therefore include the following seizure imitators^61,80,81:

i. Neurological disorders such as transient ischemic attacks, transient global amnesia, migraine, restless legs syndrome.

ii. Cardiovascular disorders such as vasovagal syncopes, orthostatic hypotension, cardiac arrhythmias, or structural heart disease.

iii. Sleep disorders such as obstructive sleep apnea, hypnic jerks, or narcolepsy.

iv. Movement disorders such as paroxysmal dyskinesia.

v. Psychological disorders such as night terror, panic attacks, or psychogenic nonepileptic seizures.